ABRALI N - Boletim da Assocracao Brasilelra do Lingtiistica,N°20, janerro de1997

BAKAIRI AND THE FEATURE ‘VOICE’

Leo Wetzels'

INTRODUCTION

lt is by now generally agreed that the feature [-voice] plays norole in the early (or lexical) phonology of any language. Recentstatements of this nature can be found in Cho (1990a, b), Lombardi(1991, 1996), lverson & Salmons (1996), among many others. Thefeature [-voice] is said to be ‘unmarked’, or, almost equivalently, isregarded as a ‘default’ feature. As such, the role of [-voice] should becon ned to the phonetic component, or, at the very most, it should beactive only in the postlexical component (cf. Lombardi 1996).Consequently, one does not expect to find a language where thefeature [-voice] must be specified at the level of lexicalrepresentation, or participates in lexical rules of any kind, includingrules of assimilation and dissimilation.

In this paper l will discuss the patterning of the laryngealfeatures of Bakairi and argue that in this language the feature [-voice]is lexically relevant. The data are taken from two studies by Souza(1991, 1995). In Souza (1995) a non-linear analysis of the Bakairifacts is given, which is different from the one proposed below inseveral respects. I will not enter into a detailed comparison betweenSouza’s analysis and my own, simply because the primary aim of mypaper is to use the Bakairi facts as an element in the discussion aboutthe lexical relevance of the feature [-voice]. Suffice it to say that, ascompared to Souza (1995), the analysis elaborated below presents, inmy view, a higher degree of analytical simplicity. Obviously, thispaper would not have been possible without Souza's studies, whichare in part based on fieldwork done by the author herself.

' Free University of Amsterdam/Holland institute of Generative Linguistics.

Leo Wetzels

1. THE D|$TR|BUT|ON OF LARYNGEAL FEATURES |N BAKA|R|

Bakairi‘ has the following system of consonantszp t k b d g

t\\‘ k\\' d\\ Q“

5 .\ I5 Z 3 'yti‘ <13

lTl Fl

ir

The syllable structure of Bakairi is of the (C)V type, whichmeans that consonants do not cluster and are only allowed in syllableonset position.

The language shows an interesting distribution of the ‘voice’feature, both within lexical roots and inside phonological words. Forboth categories, the behaviour of ‘voice’ is more or less identical. Iwill therefore construct my analysis on the basis of the distribution ofthe voice feature in lexical roots. Where convenient, mainly to justifysome analytical decisions, l will imply the word domain in thediscussion. Subsequently I will show how the distribution of ‘voice’ inphonological words can be predicted. -

1.1 Roots without Sonorant Consonants1.1.1 The data

Bakairi roots may contain sonorant and nonsonorantconsonants. Both consonant types may occur together in a root. Inthis section i will concentrate on roots which exclusively containnonsonorant consonants. in section 1.2 l will show that the patterningof voiced and voiceless consonants is not crucially different whenthey co-occur with sonorants.

1 Bakairi is an indigenous language of Brazil, and is classified as i=i llt lltllhlof the southern Carib family (see Rodrigues 198658-64) Tlin iniigiingn inspoken by approximately 350 persons who live in the strain nl M lll liiiii=i=ii_southwest of the Upper Xingu river" From Souza (199550, ftn 2) l conclude that Isl and Ill FIN! iii |i|3lllli§l|rzomplementary distribution, and that /1! niiil /3/ nie ill l-lllll|l| lt-1complementary distribution It is not cieiii liiiiii t=3niiIi=l'ti nliiily wliellmi ilivlabialised sounds must indeed be considered indupnfitjctil pliiiiioiiiec

22

BAKAIRI AND THE FEATURE ‘VOICE’

Bakairi has monosyllabic and polysyllabic roots. inpolysyllabic roots, voiced and voiceless oral obstruents alternate in away shown by the following monomorphemic words (for reasonswhich will be made clear below, I will momentarily exclude from thediscussion the root-initial consonants as in tazekada ‘bench’):

(1)

Illtazekadn ‘bench’-+

I Ip6kf)d.'t ‘woman’- +

l litubi ‘skin’+ - +

| l lodapigo ‘heat’+ ..

I ipa3ii<a ‘ant eater’

Few roots show sequences of non-alternating voicedconsonants (again disregarding root-initial consonants):

(2)

l l I Imdridri ‘jaguar’ ka[.lida 'arara’+ + +4-

l i l iriig:‘i"d.w ‘man’ azage ‘two’+ + + +-

I i l Ié"gidn ‘animal’ ezedi ‘name’

The roots of the type in (2), in which two or more consecutivenonsonorant consonants are voiced, are rare, and may be consideredexceptions. The usual distribution of voiced and voiceless consonantsconsists of an alternating sequence [+,- voice], or [-,+ voice], as in(1). Moreover, no polysyllabic roots are found which show more thana single occurrence of the feature [-voice]. in (3) l have summarised

23

Leo Wetzels

the voice patterns that one encounters in polysyllabic roots whichcontain only nonsonorant consonants.

Q)(3) ~ '*

.-~_d:i_Q_oc:r f—\l""\

.+-

_+-++++++)-++)

Although the patterns (3e-f) are not attested in thepolysyllabic roots which I have at my disposal, l suspect that theabsence of these sequences does not represent a significant fact ofBakairi phonology. Roots of the type (C)VCVCVCV, which have threeor more root-internal CV-syllables of which moreover all C's areiionsonorants, are very rare. More significantly, in derived words, forexample in words consisting of a trisyllabic (C)VCVCV root followedby one or more suffixes, the patterns of (3e-f) are normal3. Moreimportant for the proper analysis of the voice distribution in Bakairiare the patterns that are unattested for structural reasons, and whichare lacking both in roots and in words. I have summarised thesepatterns in (4a) and compared them in (4b) with the correspondingminimally different grammatical sequences.(4) a b

(unattested) (grammatical)4- - - -I-+

ll ++- +++

lll + + - +llll - +, or+ -

Clearly, the occurrence of [-voice] in Bakairi is subject totwo restrictions. The first restriction prohibits the presence of moretliiin one [-voice] feature per root. This is why the patterns (4ai, iii, iiii)are uiigrammatical. The second restriction binds an alternating [+,-] or[-,+| pattern to the first two intervocalic nonsonorant consonants of thiiroot (or to the only nonsonorant consonant of a (C)VCV root and inthe initial consonant of an immediately following suffix). In dlllmniiiturnis, the feature [-voice] can only occur on the first oi iimziiiiilliitervocalic consonant of the root (or word). This reslrlctloii Hljtl lllrithe iioiioccurrence of the pattern in (4aii).

-. .-.- .0--..-.-»o---Q-Q-Q-----1-_

Of course, the root is of the exceptional typo ('2) in words Whltrll Bliuw lliepallein (Se)

l

24

BAKAIRI AND THE FEATURE ‘VOICE’

Except for the occurrence of a single intervocalic [-voice]feature, voiced and voiceless consonants are in complementarydistribution in a way which follows from universal markednessconventions: word-initially only voiceless stops occur, whereasinteivocalically one only encounters voiced segments (except for oneposition). To clarify this point, l will turn to the root-initial consonants,which have been left out of the discussion so far.

Root-initial consonants do not take part in the game ofcontrastive voice features. To see this, compare the words in (5a)with those in (Sb):(5) a. b.

I l I Itake ‘bow ta-doka-ge ‘have a bow’

agentbow-verbaiiser-- +-

l l i lpepi ‘canoe’ i-[Jepi-re ‘his canoe’

3p-canoe-poss.- .. 4. _

l l l lpeto ‘ re’ i-[Sela ‘his re’

3p- re

As was stated before, the Bakairi syllable is either V or CV.Therefore, a root may either start with a vowel or with a consonant.For consonant-initial roots, the voice value of the consonant is alwayspredictable: when the root is not prefixed, i.e. when the root-initialconsonant surfaces in word-initial position, it will be voiceless,whereas in intervocalic position it will always be voiced. l willtherefore leave this consonant lexically underspecified and predict thephonetic values for the feature voice with the rules in (6):(6) a. Q laryngeal -) [+voice]N----V

b. 9 laryngeal -§ [-voice]/elsewhere

The morpheme structure condition in (7) formalises the factthat in root-initial position no laryngeal specifications are allowed inBakant

(7) "l<|3lar

25

Leo Wetzels

it is claimed in Wetzels (1994; in preparation) that the rules in(6) are universal markedness rules that fill in missing voice features.More in particular, it is claimed that whenever the voice opposition isneutralised in intervocalic position, the unmarked voice value is[+voice]. Very strong evidence for this fact comes from Postlexicalinteiyocalic Voicing, which, according to Wetzels (1994), may onlyoccur in languages that also have word- (or syllable-) Final Devoicing.The universal implication ‘if intervocalic Voicing then FinalDevoicing’ can only be understood properly if it can be shown thatFinal Devoicing creates the circumstances under which lnteryocalicVoicing may occur. in Wetzels (1994) intervocalic Voicing is said tobe the consequence of (a) Final Devoicing, which delinks thelaryngeal node in word - or syllable-final nonsonorant consonants, (b)a language-specific rule of Postlexical (re)Syilabification, which takesthe neutralised word-final consonants into intervocalic position, and(c), the universal markedness rules proposed in (6). which fill in themissing laryngeal specifications. it is predicted that languages thatmaintain a voice opposition word-finally (like French or English) -because they always have their laryngeal nodes specified in word-final nonsonorant consonants - could never have intervocalic Voicing.in a similar vein, there can be no language with postlexicalintervocalic Voicing which does not resyiiabify word-final consonantswith a following vowel-initial word (like Standard Dutch). Of course, avoice opposition may be neutralised by mechanisms other than FinalDevoicing, for example by a generalised rule of inteiyocalic Voice-Neutralisation, such as the Flapping rule in some dialects of Englishwhich also applies inside words, or simply because the language doesnot have a lexical voice opposition, either in general or in a specificposition. The latter situation occurs in Brazilian Portuguese, wherethere is no opposition between voiced /z/ and voiceless /s/ word-fiiially. Brazilian Portuguese does have a rule of PostlexicalSyliabification. in (8) l have exemplified the [+,- voice] alternation asit occurs in the surface manifestations of the final consonant of thelexical root /rapaS/.

(5)rapa[s] ‘boy’ inflected rapa[z]es ‘boys’

derived rapa[z]iada ‘group of boys’rapa[z]inho ‘little boy‘rapa[z]ote ‘little boy‘rapa[z]ola ‘lad’

postlexical Valteir é um ‘Valteir iii ii lively lnilnwrapa[z]energetico

26

BAKAIRI AND THE FEATURE ‘VOICE’

Clearly, the distribution of the voice values of /S/ is aspredicted by the markedness rules in (6). it is therefore not necessary.and it would even be arbitrary, to choose one of the voice values asthe lexical value. in the absence of a laryngeal node, the markednessrules of (6) will provide the appropriate voice specifications in bothintervocalic and word-final positions.

in the following section I will make a proposal about how toaccount for the distribution of the voice feature in Bakairi.

1.1.2 The AnalysisIn order to describe the Bakairi voicing facts, I have chosen

an approach which is based on a type of underspecification knownfrom tonal phonology. More in particular, l will assume that thecontrastive voice features are not linked to segmental material inlexical representation, but that they are part of the lexical entries asso called ‘floating’ features. Just like in tonal systems, we canconsequently leave the association of the voice features to theuniversal Association Convention. On the assumption. which Iconsider to be empirically motivated (cf. Odden 1994), that the voicefeature does not spread over long distances (across vowels), theAssociation Convention will associate a floating voice feature to thefirst P-bearing segment ([-sonorant] consonant) of the root. Thelexical representation of the three root types that were discussedabove are as follows:-roots which surface with a sequence [+,- voice] (cf. examples in (1)above) will be lexically represented with a floating [+voice] feature;-roots which surface with a sequence [-,+ voice] (cf. examples in (1)above) will be lexically represented with a floating [-voice] feature;-roots which surface exclusively with intervocalic voiced consonants(cf. examples in (2) above) will be presented lexically without anyvoice specification.

I have illustrated the three possible root types in (9) (theconsonants are capitalised to express the fact that they aresegmentally underspecified for ‘voice’):

(9)+s -sTIWZGKHDD TBK DD?) r~.iDnD,1

27

Leo Wetzels

The universal Association Convention associates the voicefeatures from left to right with the potentially laryngeal-nodebearing segments, i.e. [-sonorant] consonants. The language-specific constraint in (7) will block the association of the floatingfeature with a root initial consonant in words like fozekada and pekoda.Through the joint effect of the constraint in (7) and the AssociationConvention, the floating voice feature will be realised on the firstintervocalic nonsonorant obstruent. For roots which show the [-,+]pattern, the default rule (6a) will apply to fill in the missing [+voice]specifications after [-voice] has been associated. The same defaultrule will account for the words that have only intervocalic voicedconsonants without further proviso. On the other hand, the words thatsurface with a [+,- voice] pattern ask for a supplementary rule whichinsert a [-voice] feature after the lexically specified [+voice] feature. Iwill formulate the rules as in (10):(10)Voice Dissimilation Insert [-voice] / [+voice] “

Notice that it is not possible to decide on the basis of theevidence so far whether rule (10) applies before the AssociationConvention has linked up the floating voice features, or after. We willsee below, when we will discuss the behaviour of [+,- voice] in words,that the Association Convention may associate floating features toconsonants that belong to suffixes. All other positive voicespecifications will be provided by the default rules in (6). In (11) thederivation of tazekado is exemplified:

lexical rule (10)specification O

ll(11) [+voice] [-voice]

I IT 1 Z 6 K 8 D o

I I[-voice] [+voice]rule (6b) rule (6a)

Notice that this analysis, which treats the lexically relevantvoice specifications as floating features, explains why in roots thatonly contain nonsonorants, voiceless segments may only show up inone of the first two intervocalic positions. It is also not necessary to

4 Notice that this rule could be reformulated as ‘insert laryngeal node alter[+voice]‘, leaving the insertion of the negative value to the OCP, uiiiuirillyconsidered the default trigger for phonological rules (see Yip 1988).

26

BAKAIRI AND THE FEATURE ‘VOICE’

add a constraint stipulating that Bakairi allows for one [-voice] featureper word only. This fact follows from the more general restriction thatthe feature voice does not occur in the form of lexical floatingmelodies, which might in fact be a universal restriction, and OCP,which prohibits the occurrence of sequences of identical floatingfeatures. Furthermore, a unified account is given for all intervocalicvoiced consonants that are not relevant at the level of core lexicalrepresentation. Further evidence for the floating character ofcontrastive voice specifications in Bakairi will be given below insection (1.3).1.2 Roots with Sonorant Consonants

In the foregoing section I have not discussed roots whichcontain sonorant consonants. This was done for expository purposesonly. As it turns out, the analysis proposed above accountsstraightfonivardly for these roots also. Consider the examples given in(12):

(1 2) (a) imasedn ‘big’ (b) rirri31 ‘cashew’ipemugn ‘parent’ telaziéi ‘thin’iwakurn ‘pretty’ yamfi"do ‘boy’krinopiri ‘little bird’ imcmbire ‘small’

in Bakairi a sonorant consonant may be preceded or followedby a voiced, cf. (12b), or a voiceless cf. (12a), consonant. On theassumption that sonorant consonants are lexically unspecified - to myknowledge the feature [-%*-S/oice] in sonorants does not play a role inthe phonology of Bakairi - the floating feature analysis predicts that afloating [-voice] will dock onto the first available nonsonorantconsonant. in ipemugo ‘parent’, this will be the consonant thatprecedes the sonorant, whereas in imaseda ‘big’, it is the consonantthat follows the sonorant consonant. All other consonants of the rootwill surface as [+voice], as expected.

The situation is less clear in roots that contain a sequence ofa voiced nonsonorant consonant and a sonorant consonant, in anyorder. in the data provided by Souza I have found no examples inwhich a voiced nonsonorant precedes a sonorant. As for the words in(2b), where a voiced nonsonorant follows a sonorant, there is noexample where the voiced nonsonorant is followed by yet anothernonsonorant. The non-existence (or rareness?) of such forms mightbe due to the fact that quadrisyllabic roots are not very frequent inBakairi in general. In any case, the analysis thus far predicts that

29

Leo Wetzels

Voice Dissimilation should apply to such forms. Below we will see thatthis prediction is correct for derived words.

1.3 The Distribution of Voiced and Voiceless Consonants inWords

I will finally turn to the distribution of voiced and voicelessconsonants in words. To be sure, the analysis developed above onthe basis of lexical roots accounts without further proviso for thealternations of voiced and voiceless segments in words. Let us firstnotice that words which are derived from roots that consist exclusivelyof voiced nonsonorant consonants, like the ones given in (2) above.surface without any voiceless segments. This is to be expected, sincea voiceless consonant can surface only as the consequence of alexically present floating [-voice] feature, or as the result of VoiceDissiinilation. which in turn requires the presence of a lexical [+voice]spiicilicatioii. We therefore have derived ezedi-ge ‘with the name’.The same is true for roots which show a [-,+ voice] sequence: n-iik,izo-lia ‘did not ripen’. Let us next consider the words in (13), inwtiicli the last nonsonorant consonant of the root in voiceless:

(lfla) Last root consonant is voiceless

s ieka-dai object-ask-past ‘asked’ll-ti‘,/FISQ-Flgl 3pers-go-past ‘went’ipa izagi newly cleared land-to ‘to the newly cleared land’lt_].tlll-dl|€ sing~aspect she singst-ipini-ge-ba agent-food-verbaliser-negation ‘he has no food‘n epi-ge-agi object-pull-verbaliser-past ‘he pulled‘

After roots of this type. all following intervocalic nonsonorantcoiisriiiarils surface as voiced, again as expected. Now, the crucialInst case of the proposed analysis comes from words derived fromtiisyllabic roots. Strong evidence for the floating feature analysis. and,iii the same time, for the lexical presence of the [-voice] featuretZ(llllttS Iiorn words derived from bisyilabic roots of the type (C)VSV,Wllttltt S i:epreseiits a sonorant consonant. Some examples are givenin (I4)

BAKAIRI AND THE FEATURE ‘VOICE’

(14) Last root consonant is a sonorant consonant

paru-da cleared land-in ‘in the cleared land‘ila-dibe wet-participle ‘wet’n-ema-ke-agi object-hand-verbaliser-past ‘wonm-ema-ke-agi 2pers-hand-verbaiiser-past you won’s-ema-ke-G3 object-hand-verbaliser-past ‘I won‘n-ema-ge-aki objetc-hand-verbaliser-past ‘stole’

We have seen above in (12b) that the voice feature of asonorant consonant never triggers Voice Dissimilation in roots. Itwould therefore be ad hoc to explain the voiceless consonants in thesuffixes of the inflected forms of /ema/ ‘to win‘ as due to the activityof Voice Dissimilation. It is equally not true that the suffix -Ke islexically voiceless, as can be seen in the last example of (14). as wellas in many other words, like f-i'pi'ni-ge-ba ‘he has no food‘. Therefore.the root /ema/ ‘to win’ must be lexicalised with a floating [-voice]feature. interestingly, the last word of (14) clearly shows that thefeature [+voice] must be lexically present in the root /ema/ ‘to steal‘.in order to explain the presence of the voiceless consonant in thesuffix -aKi 1 Voice Dissimilation can only be triggered by a lexicallyspecified [+voice] feature. The necessity of lexical [+voice] ismoreover proved by the following words:

(15) n-ige-aki 3p-die-past ‘died’n-ike-agi 3p~sleep-past ‘slept’n-ega-tar 3p-make-past ‘made’s-eka-dai object-ask-past ‘asked’ige-ke sing-gerund ‘singing’n-ad-amge-aki 3p-intransivatortear-past ‘tore’u-di~aki 1p-go~past ‘I went’n-ige-ipie object-grate-participle ‘grated’ad—uge-ipe intransitivator-burn-participle ‘burned

The intervocalic nonsonorant consonant in a bisyilabic rootmay be voiced or voiceless. As the examples show. the value of thevoice feature of the root determines the choice for the voice value ofthe immediately following suffix-initial nonsonorant. Again. only onthe assumption that both [+voice] and [-voice] are lexical values inBakairi can we explain the alternations which occur in the suffixconsonants.

3'1

Leo Wetzels

2. SUMMARY AND DISCUSSION

In a recent paper by lverson and Salmons (1995), adistinction is made between languages like English and German,which oppose a ‘fortis‘ feature [spread glottis] (aspiration) to a ‘lenis’feature, and languages like Dutch and the Romance languages, whichoppose ‘voiced’ to ‘voiceless’. In languages of the former type,[spread glottis] is the marked feature. It is the feature that is lexicallypresent and may be active in lexical phonological processes. Thefeature [+voice] is unmarked, lexically unspecified and phonologicalinactive. On the other hand, in languages like Dutch, where it can beshown that [+voice] does spread, it is [+voice] that contrasts with theabsence of a laryngeal specification. In the light of the proposedtypoiogical distinction, it is interesting to pose the question whereBakairi fits in: is it a [spread glottis] language like English or German,or a [voice] language like Dutch or Portuguese?

There are several reasons to believe that Bakairi is not of the[spread glottis] type. First, to the best'of my knowledge, voicelessconsonants in Bakairi are not generally aspirated. Unlike in English,where different degrees of aspiration are, among other factors,related to the stress-stressless distinction, in Bakairi the (very limited)distribution of voiceless consonants is in no way related to wordstress. Stress in Bakairi words is prefinal almost without exception.Voiceless consonants typically occur as the onset of word-initialsyllables, which are usually unstressed, or are conditioned by apreceding voiced consonant, again independently of stress. Finally,voiceless consonants productively alternate with voiced consonants,as in root-initial position, where the consonant shows up as voicelessif the root remains unprefixed, but where it appears as voiced whenintervocalic. The alternation between voiced and voiceless in thesecontexts seems typical for real voice languages, like many dialects ofDutch (see Wetzels 1994), the Slavic languages, and a Romancelanguage like Portuguese (see the examples in (8) above). Also, inKorean, which has a triple laryngeal contrast between [spread glottis],[glottalised] and [voiceless], it is the voiceless consonants which arerealised as voiced consonants in inteiyocalic position.

Probably the most interesting aspect of the Bakairi data is thefact that both [+voice] and [-voice] function as lexical features.According to lverson and Salmons (1995) this should not be possible,because, if Bakairi is a [voice] language, [-voice] should not belexically present, and if it is a [spread glottis] language, [+voice]should remain lexically unspecified. The inevitable conclusion is thatthe hypothesis according to which languages that use a two-way

32

BAKAIRI AND THE FEATURE ‘VOICE’

laryngeal contrast -oppose a single laryngeal feature to the absence ofsuch a feature is incorrect.

Let me finally return to the facts of Portuguese, which,according to lverson and Salmons, is a real ‘voice’ language. Wehave seen above that Portuguese lacks a voice opposition for word-final /S/ (the only nonsonorant consonant allowed in the Portuguesesyllable coda). In order to bring out the fact that the phoneticrealisation of this consonant is predictable by universal markednessrules, it was proposed that this consonant be lexically underspecified.More in particular, /S/ surfaces as [2] in intervocalic position, whetherthis context is derived by inflectional suffixes, derivational suffixes, orpost-lexically. Consider, with this in mind, the alternations in (16):

(15)tena[s] tena[z]es (plur) tena[z]##V ‘tenacious’ tena[s]idade ‘tenacity’loqua[s] loqua[z]es (plur) loqua[2]##V 'loquacious’ loqua[s]idade ‘loquacity’feli[s] feli[z]es (plur) feli[z]##V ‘happy’ feli[s]idade ‘happiness’

The productivity of the noun-forming suffix -idade can beseen in words like simultaneidade, nasalidade, mensalidade,profanidade, irregularidade, intensidade, etc. -ldade is one of the veryfew suffixes which cause underlying /S/ to surface as voiceless inintervocalic position. It seems to me that the only reasonable way toaccount for this fact is to posit a lexical rule which fills in the [-voice]feature in root-final /S/ when followed by this specific suffix. If thisview of the fact is correct, Brazilian Portuguese provides another casefor the lexical relevance of [-voice].

References:Cho, Young-mee Yu. 1990a. Parameters of Consonantal Assimi’/ation. Unp.

Dissertation, Stanford University.

Cho, Young-me Yu. 1990b. A Typology of Voicing Assimilation. WCCFL, pp.141-155.

lverson, Gregory K. and Salmons, Joseph C. 1996. Aspiration and LaryngealRepresentation in Germanic. Phonology.

Lombardi, Linda. 1991. Laryngeal Features and Laryngeal Neutralization.Unp. Dissertation. University of Massachusetts.

Lombardi, Linda 1996. Postlexical Rules and the Status of PrivativeFeatures. Phonology 13, 1, pp 1-39.

33

Odden, David. 1994 Adjacency Parameters in Phonology. Language 70, 2,pp 289-330.

Souza, Tania Clemente C. de. 1991. The Case of Consonantal Harmony inBakairi Language (Carib). Revista D.E.L.T.A. nr. 7,1, pp 295-332.

Souza, Tania Clemente C. de. 1995. O Traco Sonoro em Bakairi. In LeoWetzels (ed.) 1995, pp 29-51.

Wetzels, W. Leo. 1994. De Fonologie van Stem. Inaugural Lecture, Deliveredat the Free University of Amsterdam, November 10, 1994. Manuscript,Free University of Amsterdam.

Wetzels, W. Leo. in preparation. Remarks on [-voice].

Wetzels, W. Leo (ed.). 1995. Estudos Fonologicos das Linguas lndigenasBrasileiras. Rio de Janeiro. UFRJ Editora.

Yip, Moira. 1988. The Obligatory Contour Principle and Phonological Rules:A Loss of Identity. Lr'nguistr'c Inquiry 19. pp 65-100.

34